!!! Now supporting with Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.1 !!!
Setup Instructions & Bug Reporting

A comprehensive Anthropic API Manifold Pipe for Open WebUI, built on the Anthropic Python SDK. It brings near-complete Anthropic Messages API parity to Open WebUI and tracks the current Claude + Open WebUI feature set: model auto-discovery, native streaming, citations, web search/fetch, code execution, Files API, Agent Skills, prompt caching, context editing, compaction, and programmatic tool calling.

Recommended Open WebUI: 0.9.0+

Features

Open WebUI Integration

• Auto-fetch Claude models and capabilities from Anthropic's Models API — no hardcoded model table babysitting
• Open Terminal-compatible on current Open WebUI releases, alongside built-in tools and MCP tools
• Prompt Caching with configurable cache scope plus CACHE_TTL (5 minutes or 1 hour)
• HTMLResponse / Rich UI tool results with embeds, files, and current OpenWebUI grouping behavior
• Notes & Channels Support fully integrated into OpenWebUI
• Task Support — Titles, Tags, and Follow-Ups
• Token + cache usage stats with per-user SHOW_TOKEN_COUNT = Off / On / With Cache
• Error reporting + retry logic for overloaded, rate-limited, and transient failures
• Admin valves for API key, custom ANTHROPIC_BASE_URL, Fast Mode, retries, timeouts, prompt caching, cache TTL, web search/fetch, programmatic tool calling, tool search, and data residency
• UserValves for thinking, thinking display, effort, token display, web limits and location, native PDF vs Files API, skills, compaction, context editing, and debug mode
• Toggle Filters for Thinking, Web Search, Code Execution, and Files API mode
• Companion Filter — routes OpenWebUI's built-in web_search and code_interpreter UI actions to Anthropic-native tools

Anthropic API Support

• Streaming with fine-grained tool streaming
• Prompt caching for tools, system prompt, and message history
• Image processing / Vision
• Native PDF upload for visual PDF analysis with charts/images
• Files API upload + download for code execution and skill workflows
• Citations for native web search, correctly placed in streamed output
• Extended / adaptive thinking with THINKING_DISPLAY = summarized / omitted / updates
• Interleaved thinking between tool calls
• Web Search with localization, citations, and dynamic filtering on web_search_20260209
• Web Fetch for direct URL retrieval
• Code Execution with persistent container reuse across turns, file upload/download, and unified code/tool/output rendering
• Programmatic Tool Calling via code_execution_20260120
• Tool Search with BM25 or Regex-based deferred tool loading
• Compaction for long-running conversations on supported models
• Context Editing for clearing old tool results and/or thinking blocks
• Agent Skills — built-in skills like pptx, xlsx, docx, pdf, plus custom uploaded skills
• Data Residency via inference_geo
• Fast Mode (research preview) on supported Opus models
• Effort parameter — low / medium / high / xhigh / max
• Tool call loop with configurable MAX_TOOL_CALLS and proactive warnings
• Parallel tool execution for local OpenWebUI tools